Developed a wedding dress recommendation engine using React, TypeScript, and Next.js, implementing flexible filtering logic to support dynamic user preferences (silhouette, neckline, style)
Architected a Neo4j-backed knowledge-graph RAG system that lets multiple specialized LLMs “trace” relationships (e.g., lace → boho style, tropical flowers → beach theme), boosting answer-relevance 10 times higher than standard vector search while preserving conversational memory across sessions
Integrated JWT-secured routing to external vendor platforms, including dynamic link generation and token signing for third-party integrations
Built responsive UI components with React Query-based async data handling and cache invalidation.
Diagnosed and resolved CloudFront network routing failures, CORS errors, and malformed token issues in production
Drove rapid prototyping to production by spinning up containerized agentic models to guide the micro-services and LLM ensemble behind Pearl Planner